Regulatory Induced Risk Aversion in Procurement Behavior: An Empirical Analysis of U.S. Investor-Owned Electric Utilities From

Transcription

1 Regulatory Induced Risk Aversion in Procurement Behavior: An Empirical Analysis of U.S. Investor-Owned Electric Utilities From Akshaya Jha December 31, 2014 Department of Economics, Stanford University, 579 Serra Mall, Stanford, CA First and foremost, I am grateful for the invaluable advice provided by Frank Wolak. In addition, comments on earlier versions of this paper by Severin Borenstein, Tim Bresnahan, Lucas Davis, Liran Einav, Jon Levin, and Steven Puller proved extremely useful as well. I thank Michael Dinerstein, Patricia Foo, and Daniel Grodzicki for feedback on multiple drafts and practice presentations. Also, I thank seminar participants from the 10 th Annual International Industrial Organization Conference, the 10 th Annual Energy Camp at the Energy Institute (Haas), as well as the GSB IO Lunch, Policy and Economics Research Roundtable (PERR) and the Third Year Seminar at Stanford. Finally, I wish to acknowledge a research fellowship awarded by the John M. Olin Program in Law and Economics in support of this work. 1

2 Abstract Electric utilities subject to output price regulation purchase much of their input coal through long-term contracts, consistently paying contract prices in excess of expected spot prices. I argue that firms purchase inputs as if they are risk averse due to the asymmetric passthrough of extreme input cost realizations into the regulated output price. I demonstrate that an expected profit maximizing firm receiving reimbursement only for costs below a threshold expresses preferences over both the mean and variance of total costs. For , I estimate a model of plant-level coal purchase, allowing for mean-variance preferences in total costs. I find that plants prefer a lower variance in input costs, and that this distortion from expected cost minimization is economically significant. As predicted by theory, the estimated magnitude of as if risk aversion is increasing in empirical measures of the regulator s propensity to reimburse extreme cost realizations. 2

3 Vertically-integrated, price-regulated electric utilities burn input coal purchased via both long-term contracts and short-term spot market transactions. These firms pay a sizable premium for contract coal over spot coal during my sample period of This contradicts the standard intuition that risk-neutral agents would make spot market purchases until the price for contract coal is equal to the expected spot coal price. Noting that average spot prices are below average contract prices in every month of my sample, why do large vertically-integrated firms enter into these higher-priced long-term contracts? The most obvious advantage of contracting is the reduction in price and supply risks relative to transacting on the spot market. However, investor-owned electric utilities 1 are typically quite large, facing a low risk of bankruptcy. Why then would expected profit-maximizing firms under output price regulation pay a premium for coal purchased from long-term contracts? I argue that the most compelling explanation is based on the structure of price regulation faced by electric utilities during this time period, which distorts the firm s input procurement decision away from expected cost minimization. 2 Namely, the regulator is tasked with setting electricity output price such that the firm can recover all prudently incurred costs; a regulator will not fully pass through extremely high spot fuel cost realizations into the firm s output price if these costs are deemed imprudent. Additionally, firms may not be fully rewarded for realizing extremely low spot fuel costs, as the regulator might lower the output price in response to these low costs. In order to avoid these adverse regulatory interventions associated with extreme cost realizations, I argue that regulated firms express a willingness to pay for a lower variance in their total costs ( as if risk aversion). To illustrate this argument, I construct a model in which an expected profit maximizing firm faces a regulated output price which does not pass through extremely high total cost realizations. I show that this optimization problem can equivalently be represented as a firm with preferences over both mean and variance of total costs. Further, I use this illustrative model to show that the magnitude of as if risk aversion exhibited is directly related to the prudence threshold above which the regulator does not pass through costs into the output price. In particular, I demonstrate that a higher prudence 1 Electric utilities can be publicly owned as well (for example, cooperative or municipal utilities). This paper focuses only on privately owned utilities facing state-level price regulation, and I will henceforth simply write utility or electric utility in reference to these investor-owned utilities. 2 Note that expected cost minimization is distinct from the fact that firms still maximize expected profits subject to the constraints imposed by regulation. 3

4 threshold (a less stringent regulator) corresponds to a higher degree of regulatory induced risk aversion. I then formulate and estimate an empirical model in which plants choose how much coal to purchase via contract versus spot markets, facing spot price and demand uncertainty. Consistent with my illustrative model of regulatory price setting, I allow plants to express preferences over both mean and variance of total input costs. I find empirically that the extent to which plants trade off the mean and variance of total procurement costs is economically significant. I also construct a simple counterfactual in which plants purchase all of their coal on the spot market, and compare the expected total costs under this scenario with the expected total costs observed in the data. This simple calculation does not rely at all on my structural model of fuel procurement, yet still provides a rough measure of the money left on the table from the regulatory distortion away from expected cost minimization. 3 For my sample of power plants over the period , the monthly average percentage increase in expected total costs relative to the spot-only counterfactual is percent, indicating further that the regulatory distortion away from expected cost minimization is sizable. 4 As fuel costs typically make up around 80 percent of a plant s variable costs (Fabrizio, Rose and Wolfram, 2007), deviations from purchasing input coal at least cost can have a substantial impact on the resulting electricity output price paid by final consumers. Moreover, I find that both the model-based and spot-only estimates of as if risk aversion are negatively correlated with variables that measure how strictly the regulator intervenes on behalf of electricity consumers (which I term regulatory stringency ) 5 ; as predicted by my illustrative model of output price regulation, plants facing less regulatory stringency (based on my empirical measures) exhibit greater levels of as if risk aversion. Regulatory interventions due to extreme cost realizations are motivated by the legal obligation regulators face in balancing the interests of electricity consumers with the 3 This calculation does rely on my estimates of expected total costs in the observed and spot-only cases; as will be described in Section 3, I estimate spot price and demand moments using within plant, time series variation in these variables. 4 In absolute terms, the monthly average over plants of this increase in expected total costs is roughly $1,726, Empirically, I create measures of regulatory stringency using data on regulatory hearings provided by the Regulatory Research Associates (RRA) division of SNL Financial (these will be described in detail below). Additionally, I compare sample period before and after the passing of the Energy Policy Act of 1992 (EPACT), as I hypothesize decreased levels of regulatory induced risk aversion after EPACT due to increased participation from unregulated suppliers of wholesale electricity generation. 4

5 interests of electric utility investors. 6 Namely, the regulator sets the electricity output price in order to allow the firm (and its investors) to recover all prudently incurred costs, while ensuring that electricity consumers do not face prices that reflect unreasonable costs. 7 The basis of my paper is testing and quantifying the prediction of as if risk aversion stemming from this legal interpretation of the regulatory environment; in contrast, the traditional approach within the economics of regulation is to think of the regulator as maximizing a weighted sum of firm profits and consumer surplus, facing agency constraints either via asymmetric information regarding some aspect of the firm s production technology or demand ( adverse selection ) and/or an unobserved action taken by the firm ( moral hazard ) 8. Therefore, this paper presents a new mechanism, based on the legal mandate to balance consumer and investor interests, by which traditional price regulation distorts firm behavior away from allocative efficiency. I also provide evidence that firms facing a more consumer-friendly regulator that disallows extreme cost realizations with higher probability (which I term a more stringent regulator) exhibit a lower degree of as if risk aversion. 9 6 Related to this interpretation, my thesis is summed up nicely by the following quotation from an industry publication: They [regulated electric utilities] need to get the buy-in of their public service commissions (PSCs), because PSCs determine what portion of the fuel cost will be paid by the ratepayer and what portion by the shareholders, Thompson said. If coal prices suddenly take a turn upward and you have elected to have a smaller contract portfolio and buy more on the spot, you re going to pay more. If you haven t gotten the buy-in of the PSC, they may say We think your plan was imprudent you can t pass that on to the ratepayers (Kasey, 2012). 7 Citing from the seminal case of Federal Power Commission v. Hope Natural Gas Co., The ratemaking process under the Act, i.e, the fixing of just and reasonable rates, involves a balancing of the investor and consumer interests. Though this case refers to the Natural Gas Act of 1938, a very similar mandate is applicable to electricity ratemaking. 8 For more detail, consider Armstrong and Sappington (2007) for a survey of agency models of regulation. Empirically, Wolak (1994) and Brocas, Chan and Perrigne (2006) structurally estimate the degree of asymmetric information facing regulators in the water utilities industry, in both cases finding economically significant evidence of information-based distortions in firm behavior. Finding a similar effect where regulation induces conservative behavior, Borenstein, Busse and Kellogg (2012) discusses how career concerns induce managers under regulation towards inaction in natural gas procurement, as action may lead to bad outcomes that reveal these managers as especially inefficient. 9 Similar to this finding, one broad conclusion reached by the empirical literature on regulation is that greater incentives (typically, the share of profit/losses accruing to the firm) decrease regulatory distortion. For example, Fabrizio, Rose and Wolfram (2007), Cicala (2012), Davis and Wolfram (2011) and others utilize the introduction of electricity generation markets in certain states (colloquially termed de-regulation ) in order to compare different outcome measures (technical efficiency, procurement costs and operating efficiency respectively) for regulated versus unregulated plants. These studies typically find greater efficiency (or lower costs) for unregulated plants relative to regulated plants. Another strand of literature examines the effects of incentive regulation, designed to reward or punish firms based on preset performance criteria ((Berg and Jeong, 1991),(Knittel, 2002)). This literature also finds an increase in technical efficiency (or a decrease in costs) associated with certain types of incentive regulation. 5

6 The remainder of the paper proceeds as follows. Section 1 describes the data, and documents the aggregate industry facts motivating exploration of regulatory induced ( as if ) risk aversion. Namely, I show that plants purchase much of their input coal via contracts, and pay a marked premium for this coal above average spot prices. In Section 2, I argue that the form of price regulation in place introduces higher order cost moments into the objective function of a regulated, expected profit-maximizing firm, using both legal evidence and an illustrative model of regulation. With this motivation, Section 3 introduces a model in which plants exhibit preferences over the mean and variance of total costs, and provides the empirical strategy utilized to estimate this model. In Section 4, I provide the results of this estimation that the magnitude of as if risk aversion displayed by plants is economically significant, as well as complementary evidence of plant-level as if risk aversion based on a comparison of the observed behavior with the counterfactual in which plants only purchase coal from the spot market. For both sets of results, I find that the degree of as if risk aversion exhibited is negatively correlated with empirical measures of regulatory stringency, confirming the prediction from my illustrative model of regulation. Finally, Section 5 concludes. 1 Aggregate Industrial Trends Motivating Exploration of As if Risk Aversion 1.1 Industrial Context Utilities own multiple generation plants, which consume different input fuels in order to satisfy demand for electricity. The focus of this paper is on the input procurement behavior of regulated coal-fired plants tasked with producing an exogenous supply of electricity. 10,11 Plants obtain input coal through two types of transactions: contract and spot. Contract purchases typically specify repeated delivery for a period in excess of one year, while spot purchases are characterized as shorter-term, typically one-time, 10 Coal-fired power plants during my sample period of were typically lower marginal cost units relative to other types of generation facilities (such as those burning natural gas), and so were tasked with producing electricity consistently throughout the year rather than turning on or off based on demand conditions. 11 For a similar plant manager interpretation of the input procurement decision (rather than a utility-level decision), see Fabrizio, Rose and Wolfram (2007). 6

7 transactions. The majority of a plant s coal is purchased through long-term contracts, with the remaining short term adjustments often conducted through spot purchases. 12 Moreover, plants typically pay a premium for coal purchased via contracts relative to spot market purchases. The primary goal of this section is to document in the data both of these aggregate industry trends regarding the large proportion of coal purchased via contracts and the contract price premium relative to spot purchases, therefore motivating an exploration of as if risk aversion in plant procurement behavior. However, we may see a contract premium above average spot prices for reasons other than as if risk aversion. In fact, previous empirical literature regarding contracting behavior in input coal procurement has focused mainly on transaction-cost economics. 13 In particular, this literature posits that relationship-specific investments such as plant location ( site specificity ), input coal mix ( physical asset specificity ), and/or reliance on a single supplier for quick delivery of coal ( dedicated asset specificity ) are important factors governing plants coal procurement behavior. 14 Then, the transaction-cost argument for the observed contract coal premium would be that higher contract prices simply reflect the higher value plants place on contract coal that is tailored to these relationshipspecific investments relative to generic coal purchased on the spot market. 15 In order to rule out this type of argument, I restrict my sample of plants to those purchasing coal from both contracts and spot markets in every month for at least 60 consecutive months (5 years) of my sample period. 16 Transaction cost explanations are unlikely to apply to these spot-inclined plants with easy access to spot coal, as the possibility of 12 Note that these contracts are legally enforceable, short of force majeure (major, unforeseeable events that cause the conditions specified in the contract to be untenable for one of the parties) or bankruptcy threats. Therefore, commitment concerns are not particularly relevant in this context, making contracts a very stable method of acquiring fuel in terms of supply risk. 13 One exception is Wolak (1996), which provides descriptive evidence that supply and price risk (among other factors) were important reasons for simultaneous participation in contract and spot coal markets. 14 Interestingly, Joskow (1987) explicitly states that It follows Williamson and Benjamin Klein, Robert Crawford, and Aremn Alchian and assumes that risk aversion is not an important factor determining the structure of vertical relationships between coal suppliers and electric utilities. This assumption demonstrates the tension between differentiating risk aversion and transaction cost effects; namely, in Joskow s case, allowing for risk aversion may complicate the interpretation of his results under asset specificity. 15 For a concrete example of this argument, the coal purchased from contracts may adhere more closely to the plant s optimal input coal mix than the composition of coal available on the spot market. For my sample, I can refute this argument, as the average observable characteristics (Btu content, ash content, sulfur content) of coal purchased from contracts versus the spot markets are very similar. 16 This sampling scheme results in plants located primarily in the East and Midwest, where spot markets are more prevalent. 7

8 ex-post holdup by the mine is small. 17,18 The fact that these plants still contract therefore seems more likely to be attributable to as if risk aversion rather than transaction cost explanations. A more mechanical argument for why we observe a premium for coal purchased via contract relative to spot market coal is simply that contract prices adjust more slowly than spot market prices to economic factors affecting coal production. Under this claim, in periods of declining coal production costs, we would see contract prices fall at a slower rate than spot prices, leading to the observed contract premium. However, note that the majority of contract prices are indexed in some form, making it less likely that the contract price premium (over spot purchases) is due to older contracts signed prior to a period of declining fuel prices. 19 Furthermore, this partial adjustment argument would still not explain why plants are consistently signing long-term contracts at higher prices; it is highly unlikely that plants would repeatedly over-estimate the expected future spot price when signing these higher priced contracts. Finally, note that the majority of contracts only specify annual upper and lower bounds on the quantity that must be purchased from that contract; plants have significant flexibility in how much they buy each month from existing contracts. The question remains as to why plants still choose to purchase the majority of their coal from these (higher priced) contracts. I argue that, for my sample of plants that purchase regularly from both contracts and the spot market, the observed contract premium reflects plant preferences for contract coal due to regulatory induced risk aversion. Namely, I argue in the next section that coal purchases from existing contracts face much less regulatory scrutiny relative to purchase from the spot market. This paper sets out to recover the magnitude of as if risk aversion implied by the observed monthly purchases from typically higher priced, but certain, 17 Recall that transaction cost theories apply when the possibility of ex-post hold up in repeated bargaining between plant and mine is significant in the absence of ex-ante contracting arrangements. For one example, a coal mine may have significant bargaining power over a plant that is located near it; a contract between mine and plant signed before the plant is built at a given location is likely to mitigate this bargaining power. 18 Joskow (1985) argues similarly that transaction costs are unlikely to be important for plants with easy access to multiple suppliers of coal. Within that paper s framework, the plants in my subsample would fall into Case 1 and Case 2, characterized primarily by flexibility in potential suppliers of their input coal. 19 From Joskow (1988), which describes these contract pricing practices in more detail, the typical long-term coal supply contract thus uses a BPE pricing formula in which an initial base price is set when the contract is negotiated and then adjustments are made to the base price using a weighted average of changes in external input price, productivity indexes, and changes in actual costs. 8

9 existing contracts versus lower priced in expectation, but uncertain, spot purchases. To do so, I more formally derive a model of plant demand for contract versus spot coal in Section 3. Regarding coal supply, I model the plant as taking both contract and spot prices as given when making its procurement decisions. For contract prices, this reflects the fact that bargaining over coal contracts takes place in the past; therefore, prices for existing contracts are pre-determined when the plant decides what level of quantity to purchase from these contracts in each month. For spot prices, recall again that I restrict myself to a sample of plants that purchase spot market coal in every month of sample for at least 60 consecutive months. Therefore, these plants have ready access to a fairly liquid spot market for coal; moreover, coal suppliers have many alternative buyers to choose from. As a result, for my subsample of plants, the amount of spot coal one plant purchases is unlikely to substantially shift the spot market price it faces for this coal Data Sources and Construction The primary source for the data used in this analysis is FERC Form 423 maintained by the Energy Information Administration (EIA). This dataset provides transaction-level data on the month and year of purchase, type of transaction (contract versus spot), quantity received, price, fuel type (ex: coal, natural gas, oil), and various fuel characteristics such as British Thermal Units (BTU), sulfur, and ash content. 21 I also construct measures of how strictly the regulator intervenes on behalf of consumers ( regulatory stringency ) from data provided by SNL Financial; these variables are described more fully in the next section regarding the regulatory structure in place during my sample period. All summary statistics are relegated to Tables A1 and A2. 22 constructed the data is found in the Data Appendix. A full description of how I 20 Given that coal deliveries being classified as contract versus spot does not affect coal suppliers underlying production and transportation costs, the existence of a contract price premium for plants with easy spot market access is evidence that some party in the supply chain (either coal mines or coal transportation) can exercise market power over buyer (read: plant) risk preferences. However, as I am interested in demand side parameters (i.e: plant risk preferences), I do not explicitly model this bargaining process (or coal supply more generally) for this paper. 21 The EIA only provides fuel cost data for regulated plants. For this reason, I cannot perform an explicit comparison of the input procurement behavior of plants in my sample with plants that are not under rate of return regulation. Also, I choose to end my sample period in 1998 as an increasing number of plants were divested to unregulated firms after this period (and so would drop out of my sample). 22 Any tables and figures marked with an A are in Appendix A. 9

10 1.3 Aggregated Time-Series of Contract Premium and Contract Proportion In Figures 1 and 2, I show the time series from of the quantity-weighted average contract price premium over spot purchases 23 and aggregate proportion of coal purchased via contracts respectively. This variation comes directly from the raw FERC Form 423 data, which comprises all deliveries made to utility-owned power plants with capacity of at least 50 megawatts. 24 Examining Figure 1, we see the puzzling fact that there is a consistent average premium paid for coal that is purchased from contracts relative to coal that is purchased on the spot market in the same month. Moreover, in the face of this premium, Figure 2 tells us that coal-fired power plants still purchase the majority of coal from these higher priced contracts. It is exactly this monthly variation in the contract price premium facing plants, and the resulting variation in how much is purchased via existing contracts versus a spot market with uncertain price, that I use to estimate the degree of as if risk aversion displayed in input procurement behavior. In Figure 1, we see that the average contract premium dips in 1992, and again beginning in From Figure 2, we see a decline in the aggregate proportion of coal purchased via contracts in As I will describe more fully in the next section regarding regulatory structure, the Energy Policy Act (EPACT) of 1992 opened the national transmission grid to wholesale electricity suppliers, which reduces the potential effects of regulatory-induced risk aversion both through the entry of these suppliers into the electricity market as well as the increased opportunity for regulated utilities to purchase electricity from these generators if their own costs of production are sufficiently high. EPACT was subsequently executed by FERC beginning in Therefore, these simple graphs provide preliminary descriptive evidence that the contract price premium indicative of plant as if risk aversion is brought about by the regulatory structure in place In Appendix Figure A.1, I plot the contract and spot price series separately for the same time period, in order to give a sense of how large these differences are relative to the price levels. 24 The trends described in this subsection are extremely robust. For example, similar findings emerge if I create these figures for a balanced panel of plants transacting in contract and spot markets in every month of the sample (I also balance plants using a variety of other time frames and find the same trends). In addition, there are not observable differences in quality (read: sulfur, ash, moisture, Btu content, etc.) for coal delivered by contract versus spot transaction types; thus, coal quality is very unlikely to explain the contract price premium. 25 After my sample period ends in 1998, the contract premium diminishes even further. In fact, the average premium (calculated in the same manner as Figure 1) is zero by the middle of the year

11 Figure 1: Contract Price Premium over the Spot Coal Price: Quantity-weighted Mean Contract By Premium Month, over Spot Prices Quantity-weighted Averages over Plants for :36 Wednesday, July 3, Contract Price Premium ($/million BTU) JAN87 01JAN88 01JAN89 01JAN90 01JAN91 01JAN92 01JAN93 01JAN94 01JAN95 01JAN96 01JAN97 01JAN98 01JAN99 Month of Sample Notes: This graph is created using quantity-weighted averages of both contract and spot prices over all coal deliveries in the full FERC Form 423 sample for each month. FERC Form 423 reports delivered prices of coal (FOB plant), including transportation costs. The definition of contract used in FERC Form 423 is an agreement to purchase input coal, with repeated deliveries, lasting greater than one year. All other deliveries would be considered spot purchases. Sources: The contract and spot prices are obtained from FERC Form 423 data on all purchases made by electric utility owned power plants with nameplate generating capacity of at least 50 megawatts. 11

12 23:36 Wednesday, July 3, Figure 2: Proportion of Coal Purchased Via Contract, Proportion of Coal Purchased From Contracts Aggregated Proportion over Plants for Contract Proportion JAN87 01JAN88 01JAN89 01JAN90 01JAN91 01JAN92 01JAN93 01JAN94 01JAN95 01JAN96 01JAN97 01JAN98 01JAN99 Month of Sample Notes: This monthly proportion is created by summing the quantity purchased over all coal deliveries from contracts and dividing by the sum over the quantity purchased over all coal deliveries (including spot purchases) in the full FERC Form 423 sample for each month. The definition of contract used in FERC Form 423 is an agreement to purchase input coal, with repeated deliveries, lasting greater than one year. All other deliveries would be considered spot purchases. Note that the y-axis does not intersect the x-axis at zero. Sources: The contract and spot quantities are obtained from monthly FERC Form 423 data on all purchases made by electric utility owned power plants with nameplate generating capacity of at least 50 megawatts. 12

13 2 Examination of the Regulatory Structure 2.1 Overview In this section, I set out to describe how the regulatory structure in place during my sample period induces electric utilities to act as if they are risk averse in their input procurement. In particular, I argue that the regulatory disallowance of imprudently high fuel costs, as well as the potential for regulatory clawback of excessively high profits (low costs), creates incentives for regulated firms to avoid these extreme fuel cost realizations. This argument is formalized by an illustrative model in which an expected profit maximizing firm faces a regulated output price that passes through (read: reimburses) only realized costs below a prudence threshold. I show that the objective function of this expected profit maximizing firm under price regulation can equivalently be expressed as an unregulated firm that minimizes a weighted sum of their mean and variance of total costs ( as if risk aversion). Further, this model generates the comparative static that a lower prudence threshold (higher regulatory stringency) corresponds to a smaller distortion away from expected cost minimization (lower degree of as if risk aversion). Based on this, I describe the empirical measures of regulatory stringency used to test this model prediction in the data. 2.2 Description of the Regulatory Structure During my sample period of , the vast majority of electricity generation was produced under output price regulation. Under this type of regulation, the state regulatory commission 26 is tasked with setting output electricity price in order to allow the utility an opportunity to recover its total costs 27, including a return on capital invested. In setting the output price, there is a distinction between longer term costs such as plant construction and investment and shorter-term, more variable operating and maintenance (O & M) expenses such as fuel procurement. The longer term costs are included in a rate base and, after one review determining the prudency of the investment, are 26 Or the Federal Energy Regulatory Commission (FERC), in federal cases where the actions of the utilities or their holding companies affect multiple states. 27 From the case of the Federal Power Commission v. Natural Gas Pipeline Co.: regulation does not insure that the business shall produce net revenues. 13

14 amortized into the output price set by the commission. Traditionally, operating expenses such as fuel procurement costs 28 are carefully scrutinized during irregularly scheduled rate proceedings, and the commission s decisions provide guides during the interim periods between these proceedings. 29 These operating expenses make up an average of 59 percent of operating revenues, indicating that changes in input procurement costs have substantial effects on the output electricity price faced by consumers. 30 Deviating from traditional rate setting proceedings due to rapidly rising fuel costs in the 1970s, many state commissions approved the use of fuel adjustment clauses (FACs). 31 In particular, as firms facing these rapidly rising costs constantly initiated rate reviews asking for output price increases, FACs were designed to pass these fuel price increases directly to electricity consumers without the processing lag 32 associated with formal regulatory meetings. In the FAC case, realized fuel costs are passed directly into the output price via a preset formula; absent a FAC, the state regulatory commission typically sets an output price before input purchases are made intended to allow the firm an opportunity to recover these fuel costs. 33 However, state commissions employing FACs almost universally schedule periodic rate proceedings (for example, annual meetings) in order to review the operating costs passed through via the FAC. Any input costs deemed to be imprudently passed into the output price are recouped in these proceedings. Moreover, Schmidt (1980) notes that some FACs in practice exercise partial passthrough, in which 100 percent of the cost decreases are automatically passed through, but less than 100 percent of the cost 28 Note that, for electric and gas utilities, fuel procurement costs represent between 50 and 60 percent of operating expenses. This statistic is cited on page 260 of the Third Edition of Phillips and Brown (1993). 29 This description of the regulatory treatment of operating expenses is also from page 260 of the Third Edition of Phillips and Brown (1993). 30 This statistic is from (Phillips and Brown, 1993), page 255 of the Third Edition. 31 The majority of states are under some form of fuel adjustment clause during my sample period. The table of states under FAC regulation over time is available upon request. 32 The time between filing a rate case and adjudicating this case was often in excess of a year. 33 One argument made in favor of FACs is that firms cannot materially affect the fuel prices faced through better procurement strategies; under this claim, firms are price-takers in the coal market (see (Graves, Hanser and Basheda, 2007) for further exposition regarding this argument). However, plants can affect the coal price they face via the choice of contract versus spot coal; this contract/spot decision has not, to my knowledge, been addressed in the debate regarding FAC utilization. 14

15 increases are passed through to the consumer. 34,35 These types of prudency concerns are also relevant in the less common rate proceedings case; a commission can still disallow input costs ex-post that it deems to be imprudently incurred. 36,37 In sum, I claim that these prudency concerns arising in both types of price setting methods (FACs and rate proceedings) result in the regulatory punishment of extremely high cost realizations, which in turn induces firms to take action in order to avoid these extreme realizations. Finally, note that the regulatory review of contracts as it relates to prudency may inherently be less strenuous than that faced by spot purchases. If a contract was approved by the commission at the time of signing, coal purchases from this contract are typically more difficult to deem imprudent at a later point in time than simple spot purchases. 38 Therefore, from a regulatory standpoint, we observe a contract premium over spot prices for two related reasons: (1) contract purchases have less price and supply risk than spot purchases 39 and (2) for a given level of price and supply risk, purchases from contracts are less likely than spot purchases to be deemed imprudent because the commission previously approved the signing of the contract. 34 This type of partial passthrough is designed to enhance incentives for cost minimization with respect to fuel procurement, and so comes under the heading of incentive regulation. Further details regarding incentive regulation programs can be found in Joskow and Schmalensee (1986), Berg and Jeong (1991), and Knittel (2002) (among others). 35 Graves, Hanser and Basheda (2006) argues that this asymmetric passthrough may also be a feature of traditional rate-setting meetings. The authors quote is: The risk of regulatory disallowance of a large deferred balance of fuel or purchased power costs (absent an AAC) does not involve both up and downside prospective returns for investors. Rather, it only involves a downside possibility. Note that AAC is an abbreviation for automatic adjustment clause, which is synonymous with FAC. 36 From Phillips and Brown (1993) on page 260, Operating expenses may be controlled in two broad ways, by disallowing improper charges already incurred in rate proceedings, and by prohibiting extravagant or unnecessary charges before they are incurred. 37 Legally, prudency reviews are supposed to determine if a reasonable manager at the time of the decision would have purchased inputs similarly to the actual procurement choice, but this ideal can be difficult to follow ex-post. For example, in the case of Delmarva Power & Light v. the Public Service Commission of Delaware, the appeal states Additionally, Delmarva complains that the Commission used hindsight judgment as a standard for determining the prudence of such expenditures. 38 Commissions have in rare cases disallowed contracts ex-post that reflected unreasonably poor managerial judgment. However, in a paper in the legal field describing some of these cases, Templeton (1986), says that once a regulatory agency has blessed a particular contract provision or set of terms, coal companies may find utilities extremely reluctant to deviate from these terms for fear of incurring renewed regulatory criticism. 39 Consider this example, again from the Supreme court ruling for City of Newark v. Delmarva Power and Light Co: Delmarva had also concluded that to be assured a dependable supply of compliance coal for its new plant, two-thirds of its annual tonnage should be obtained under long-term contract, with the balance through the spot market, or shorter term contracts. 15

16 2.3 Ex-post Punishment of Extreme Cost Realizations Induces As If Risk Aversion From the Federal Power Commission v. Hope Natural Gas Co, the following quotation is often cited: it is the result reached not the method employed which is controlling in determining whether a given output price set by the regulator should be legally upheld. However, another important finding from the same case is that the fixing of just and reasonable rates involves a balancing of investor and consumer interests. Namely, in assessing the legality of the result reached by regulation, a range of output electricity prices is reasonable based on differing interpretations of how to balance the competing interests of utility investors and electricity consumers. 40 In practice, this creates a regulatory environment in which the status quo is maintained as long as the profits earned by the firm are not too high or too low. 41 Based on this environment, the regulation in place induces bounds on realized total costs 42 as depicted in Figure 3 below. 43 Examining Figure 3, there are three cases to consider: 1. Realized firm profits corresponding to the range of total costs [T C L, T C H ] are considered reasonable by the regulator. In the case where fuel costs are passed through via rate proceedings, the firm earns the profit associated with these realized costs, noting that the output price is set by the regulator ex-ante. 44 In the FAC case, total costs in [T C L, T C H ] are fully passed through into the output price; they are not incurred by the firm (the firm earns zero profit associated with these fuel costs). 40 For further evidence, consider this quote from the Pennsylvania Commission: There is a range of reasonableness within which earnings may fluctuate and still be deemed just and reasonable and not excessive and extortionate. This quote is on page 382 of the Third Edition of Phillips and Brown (1993). 41 Joskow (1974) describes a similar effect stemming from an economic and political structure in which the actions of a passive regulator induce implicit profit bounds for the firm. Briefly, the nominal output price is the politically salient object for the regulator; as long as output electricity price remains constant or is decreasing, the constituency will tend to be happy with the performance of the commission. As a result, firms will only file for price increases when rising costs make this action absolutely necessary. 42 Note that I consider costs rather than profits because the revenue stream accruing to the firm is fixed via regulation. 43 Examples of regulatory theory papers regarding electric utilities which analyze these lower and upper profit bound type models in a static setting include Burness, Montgomery and Quirk (1980), Isaac (1982), and Braeutigam and Quirk (1984). 44 In the case where the realized profits corresponding to total costs in [T C L, T C H ] are deemed too low by the regulator, the firm will receive an output price increase as these costs are in the prudent range. 16

17 Figure 3: Regulatory Recompense by Realized Costs pdf Exorbitance Bound: TC L Prudency Bound: TC H TC Notes: This graph is a visual aid for the argument as to how the regulatory process in place can induce as if risk aversion. For this graph, I assume that revenue is fixed by regulation, and that total costs are normally distributed. The yellow portions of the graph are ranges of costs where the firm incurs some form of regulatory punishment (described in the text), whereas the unshaded area corresponds to the reasonable range of incurred costs as perceived by the regulator. 17

18 2. Next, let us consider total costs less than T C L. This realization corresponds to profits which are too high from the regulator s perspective. For costs in this range, consumer advocacy groups may complain that the firm is receiving excessive profits at the consumers expense and force the commission to hold a rate-setting meeting in order to lower the electricity output price. 45 Under a FAC, realized costs (however low they may be) are typically passed through directly into the output price; however, as with rate proceedings, these lower costs may be used as evidence to adjust downwards the formulaic reimbursement received for future costs In practice, the more relevant regulatory intervention to consider is when the total costs incurred are above T C H. 47 Total costs in this range are considered imprudently incurred, and may be disallowed by the state commission. 48 In both the rate proceedings and FAC cases, total costs in this range would not be passed into the output price, and would instead be incurred by the firm itself. Under this setup, it is clear that a firm will optimize over higher moments of cost, rather than simply minimize their mean costs. In particular, an expected profit maximizing firm facing these regulatory constraints may, for example, try to reduce the variance of their costs in order to avoid the consumer group scrutiny implied by realized costs below T C L and the risk of regulatory disallowance associated with realized costs above T C H. This regulation-based distortion away from expected cost minimization is what I term as if risk aversion. This distortion is induced by the punishment of extreme cost realizations used in practice by regulators in order to balance investor and consumer in- 45 Braeutigam and Quirk (1984) provides some descriptive empirical evidence that firms actually preempt consumer initiated reviews by asking for rate decreases when their profits become high enough to warrant consumer group scrutiny. 46 For one recent example of this, in both 2011 and 2012, the Georgia Public Service Commission (PSC) has decreased rates paid to Georgia Power in response to lower fuel costs. In 2011, the consumer group Georgia Watch claims to have formally intervened during rate proceedings on behalf of consumers (Georgia Watch, 2011). In 2012, Georgia Power itself filed with the PSC for a rate decrease, preempting consumer group pressure (Swartz, 2012). 47 Note that we empirically see more rate cases in periods with increasing costs (see, for example, Joskow (1974)). From the SNL rate case data (described below), we see that the vast majority of rate cases are for rate increases, and very often, the rate increase granted by the state commission is less than that requested by the firm (which speaks to the regulator s prudency concerns). 48 One example, from the City of Newark v. Delmarva Supreme Court Decision: It knew, or should have known, that the spot market for coal was volatile and shallow, particularly after a 72 day nationwide miners strike. While Delmarva had no way of knowing that spot prices would rise dramatically, there was no requirement that it waive its right to additional deliveries at the time it did.... In exercising the waiver it assumed an unnecessary risk and placed upon its ratepayers the cost of guessing wrong. Such conduct was clearly imprudent and the Commission finding to that effect must be sustained. 18

19 terests. Unlike traditional agency based models of regulation, I am not forced to assume anything regarding the regulator s underlying preferences, information structure, and so on in empirically identifying this distortion. 2.4 Illustrative Model of FAC-induced risk aversion To more formally illustrate that the regulatory punishment of high cost realizations can induce as if risk aversion, I construct a model in which an expected profit maximizing firm under regulatory constraint no longer minimizes expected total costs; instead this firm minimizes a weighted sum of their mean and variance of total costs. 49 In this formulation, the firm faces random total costs T C which are normally distributed with mean µ and variance σ 2 (µ). I assume that σ 2 (µ) is a decreasing function µ; this ensures that choosing a lower mean for the distribution of total costs results in a higher variance in total costs. Then, the firm maximizes expected profits over the parameter µ M, subject to the constraint that total revenues R(T C) are set ex-post by the regulator such that all realized total costs below a given prudence threshold T C are reimbursed. This results in the following maximization problem for the firm: max µ M subject to E[R(T C) T C] R(T C) = { T C if T C T C T C otherwise (1) Using the fact that T C Normal(µ, σ 2 ), we know that 50 : 1. Probability that realized costs are deemed imprudent: p P rob[t C > T C] = 1 Φ( T C µ σ ) 2. Expected Profits: E[0 T C T C](1 p) + E[T C T C > T C]p = (µ + σ Finally, I note that dp dt C φ( T C µ σ ) p )p 51 T C µ = φ( )/σ, and show that the above expected profit σ maximization subject to FAC pass-through regulation (in Equation 1) can be approxi- 49 Full derivation of this result is relegated to Mathematical Appendix B; a sketch of this derivation is provided below. 50 I denote Φ(t) as the standard normal cumulative distribution function and φ(t) as the standard normal probability density function. 51 The equation for E[T C T C > T C] is given by the truncated Normal distribution, typically utilized in sample selection or censoring models. 19

20 mated by an unregulated firm minimizing mean and variance of total costs (Equation 2 below): 52 min µ M E[T C] 1 dp V ar[t C] min p dt C µ M µ + λσ2 (µ) (2) Namely, an expected profit maximizing firm facing regulation with partial passthrough does not minimize total expected costs; instead it exhibits mean-variance preferences over total costs. In particular, the firm is willing to trade off a higher mean for lower variance of costs in order to lower the amount of mass residing above the prudency threshold where costs are not reimbursed. 53 Moreover, I show that the firm s willingness to trade off mean and variance in costs (λ 1 ) is an increasing function of T C.54 p dt C In words, this illustrative model predicts that, all else equal, a less stringent regulator who only disallows realized costs that are very high will induce more distortion in firm behavior away from expected cost minimization (the firm will place a low weight on the variance of total costs in optimization). Therefore, if my hypothesis of regulatory induced ( as if ) risk aversion is correct, we should expect empirically that my quantification of the degree of risk aversion exhibited in coal procurement is negatively correlated with measures of regulatory stringency. However, note that this illustrative model is not meant to perfectly capture all elements of the regulatory structure in place, but rather to show the economic intuition behind the claim that an expected profit maximizing firm under regulatory constraints can exhibit preferences over higher moments of cost as well. Based on this motivation of regulation, I impose mean-variance preferences in total input costs in the fuel procurement model specified in the next section. dp 2.5 Covariates Associated with Regulatory Stringency In order to assess how risk aversion in fuel procurement varies with regulatory structure, I utilize both state-level variation in regulatory regime and the policy changes enacted 52 The approximation relies on the claim that the firm s choice of µ (which determines σ 2 (µ)) primarily affects their expected profits through the change in E[T C T C > T C]; the change in p, the probability of the prudence constraint binding, from a change in µ must be second order. Numerical justification of this claim is provided in Mathematical Appendix B. 53 Note that we are assured that λ 1 dp dp T C µ p is greater than zero, as = φ( dt C dt C σ )/σ < 0. This condition ensures that the firm is as if risk averse rather than risk loving. 54 dλ I show in Mathematical Appendix B that > 0 over the relevant range of T C. dt C 20